Has anyone else noticed a sudden lack of capacity for G6 and G7 GPU instances in eu-central-1? I've used these instance types regularly without problems, but this morning a wide range of them started returning capacity errors. Other regions, including us-east-1 and us-west-2, still appear to have availability. Is this a regional outage, unusually high demand, or a broader capacity issue?
3 Answers
High demand from LLM training and inference could definitely be contributing. G6 and G7 instances use L4 GPUs, which are also useful for graphics and rendering workloads, so several types of customers may be competing for the same limited hardware.
I’m seeing the same capacity error in eu-central-1, while us-east-1 and us-west-2 are working normally. That points more toward a regional capacity shortage than a general problem with the instance types.
The most likely explanation is simply that the region is temporarily full. GPU demand has grown sharply because of AI and machine-learning workloads, and providers can’t always add new hardware quickly when equipment and data-center capacity are constrained.
What makes you think our own LLM usage is a significant factor here? The regional pattern seems more important than any one customer’s workload.